The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/watch.py invokes external commands using subprocess.run to interact with the agent-browser CLI and verify the presence of npm. On Windows systems, it employs shell=True to ensure compatible command resolution within the shell environment.\n- [PERSISTENCE]: The skill programmatically updates the ~/.agent-browser/config.json configuration file to set autoConnect: true. This modification is intended to ensure that the automation tool consistently recognizes and connects to the user's active browser session during subsequent runs.\n- [INDIRECT_PROMPT_INJECTION]: The skill acts as a surface for indirect prompt injection by ingesting external data from user-provided files and scraping content from live Twitter web pages.\n
Ingestion points: The agent reads tweet links from a user-specified text file (links.txt) and extracts interaction data (likes, retweets, etc.) from the HTML content of Twitter pages.\n
Boundary markers: Explicit boundary markers or instructions to ignore embedded commands are not utilized, as the data processing is performed by a specialized Python script rather than being directly interpolated into an LLM prompt.\n
Capability inventory: The skill possesses capabilities to execute shell commands, write to the local file system (for reports and configuration), and perform network operations via the controlled browser.\n
Sanitization: The script implements regex-based extraction for tweet IDs and interaction metrics, effectively filtering out non-numerical content and limiting the risk of malicious text influencing the application logic.

twitter-watch